DeepSeek has launched the developer preview of DeepSeek Harness (DSH), an open-source framework designed to help developers build and manage autonomous agents. Built upon the Cordis plugin system, the architecture is entirely modular, allowing developers to swap or recompose core capabilities—including models, tools, skills, sessions, sandboxes, storage, loops, scheduling, and the UI—via configuration without modifying the underlying source code. A key technical feature is the system's comprehensive traceability; every interaction is captured in an append-only session log that records system prompts, reasoning steps, tool calls, results, subagent scheduling, and context injections. Through a specialized Trajectory view, users can inspect these records and perform operations such as resuming, forking, searching, or replaying the event stream. The framework supports four distinct runtime modes: Standard mode for full toolsets; Code mode for orchestrating multi-round tool calls via model-generated code; Minimal mode for benchmarking models using only a shell and file editor; and Creator mode for testing Cordis plugins in memory. Developers can deploy the Web UI quickly using "npx @deepseek-ai/dsh web" or install the full source code from the official GitHub repository to participate in the evolving plugin ecosystem.